Discriminate between datasets and scalars in DatasetSchedule

[javihern98]

Initial Checks

• I have searched existing issues for duplicates

Problem Description

DatasetSchedule.global_inputs is a flat list of all external dependencies without distinguishing datasets from scalars. This causes two problems:

1. Imprecise validation: _validate_extra_datasets checks datasets against all global_inputs, which includes both datasets and scalars mixed together (to be introduced in Validate that data_structures does not contain datasets not referenced by the script #569)
2. Missing scalar tracking: Scalars used inside RegularAggregation (calc, filter, rename, etc.) end up in unknown_variables and are never included in global_inputs at all — a gap in dependency tracking.

Current behavior:

from vtlengine.API import create_ast
from vtlengine.AST.DAG import DAGAnalyzer

script = """
    SC_r := SC_1 + SC_2;
    DS_r <- DS_1[calc Me_2 := Me_1 + SC_r];
"""

ast = create_ast(script)
ds = DAGAnalyzer.ds_structure(ast)

print(ds.global_inputs)  # ['DS_1'] — SC_1 and SC_2 are missing!

Proposed Solution

Split global_inputs into four categories in DatasetSchedule, with global_inputs as their union (no duplicates):

Field	Context	Behavior
global_input_datasets	Definite datasets (dataset operand of RegularAggregation, Identifier with kind="DatasetID", feeds into dataset ops)	Must provide a dataset
global_input_scalars	Definite scalars (feeds only scalar chains from constants, identified via propagation)	Must provide a scalar
global_input_dataset_or_scalar	Top-level VarID ambiguity (e.g., DS_r <- X + 2 where X could be dataset or scalar)	User may provide either
global_input_component_or_scalar	Inside RegularAggregation ambiguity (e.g., DS_1[calc Me_2 := Me_1 + X])	User may provide a scalar, but semantic error 1-1-6-11 is raised if it collides with a component name

Expected behavior:

from vtlengine.API import create_ast
from vtlengine.AST.DAG import DAGAnalyzer

script = """
    SC_r := SC_1 + SC_2;
    DS_r <- DS_1[calc Me_2 := Me_1 + SC_r];
"""

ast = create_ast(script)
ds = DAGAnalyzer.ds_structure(ast)

print(ds.global_input_datasets)            # ['DS_1']
print(ds.global_input_scalars)             # ['SC_1', 'SC_2']
print(ds.global_input_dataset_or_scalar)   # []
print(ds.global_input_component_or_scalar) # []
print(ds.global_inputs)                    # ['DS_1', 'SC_1', 'SC_2']

Classification approach (pure AST analysis):

1. Track which statements involve dataset operations (has_dataset_op flag) — set when visiting RegularAggregation, JoinOp, Aggregation, Analytic, HROperation, DPValidation, or Identifier with kind == "DatasetID".
2. Identify scalar outputs via fixed-point propagation: seed from constant-only assignments (a := 1), then propagate through statements with no dataset ops where all inputs are known scalars.
3. Classify global inputs using the decision tree:
   • In unknown_variables (from RegularAggregation) → global_input_component_or_scalar
   • Used as input to a statement with has_dataset_op=True → global_input_datasets
   • Feeds ONLY into scalar chains (no dataset ops) → global_input_scalars
   • Otherwise → global_input_dataset_or_scalar

Consumers updated:

• _validate_extra_datasets → use global_input_datasets + global_input_dataset_or_scalar
• _extract_input_datasets → return global_input_datasets + global_input_dataset_or_scalar
• _save_datapoints_efficient → use global_input_datasets

Alternatives Considered

• API-layer classification: Cross-reference global_inputs with loaded datasets/scalars dicts after both DAG analysis and data loading. Simpler but doesn't fix the missing scalar tracking gap (scalars in RegularAggregation never appear in global_inputs).
• Two categories only (global_input_datasets + global_input_scalars): Doesn't capture the ambiguity cases where the user can provide either type, or where component/scalar collision detection is needed.

Additional Context

Key files:

• src/vtlengine/AST/DAG/_models.py — StatementDeps, DatasetSchedule
• src/vtlengine/AST/DAG/__init__.py — DAGAnalyzer, _ds_usage_analysis()
• src/vtlengine/API/__init__.py — _validate_extra_datasets, _extract_input_datasets
• src/vtlengine/Interpreter/__init__.py — _save_datapoints_efficient